An upside down snow-globe and a hovering hat
How Lilly learned to offer movement, rather than wait to be moved
Some learners cannot be guided by words, modeled movements, or a teacher’s hands. For them, every new skill must begin from scratch, with the environment arranged so that movement starts on its own and is shaped step by step. This is the case with Lilly, an eight-year-old with no discriminative identity match-to-sample or imitation repertoire and a long history of physical prompting.
Her story begins with an upside-down snow globe and a hovering hat. But it is also about something much larger: what B. F. Skinner once described as the true aim of education: building the repertoire that remains when specific responses have faded, the ability to learn.
A pause and an upside-down snow globe
Lilly, aged eight, sits across from her mum, watching an iPad while her mum gathers materials. Mum says, “Pause.” Lilly immediately places her hands flat on the table and orients to Mum. Mum presses pause on the device, and Lilly hands the iPad to her. Lilly then places her hands back on the table and reorients. Mum sets the iPad on the floor to the side. She places a snow globe upside down on the table and positions her hands on the table, that’s Lilly’s signal to start her own movement. Lilly rotates the globe upright. The instant it is turned, Mum says “yes.” Lilly releases the globe, repositions her hands flat on the table, and reorients. The iPad is delivered.
A hovering hat
About 30 seconds later, the reinforcement release sequence begins again. This time, Mum holds a hat in a hovering position about 5 cm above Lilly’s head, without touching her head. Lilly looks up. Mum says “yes,” removes the hat, and starts the reinforcement delivery chain.
On the next trial, the hat is again held in a hovering position about 5 cm above Lilly’s head. Lilly looks up, Mum says nothing. Lilly then lifts her arms and touches the hat with both hands. Mum says “yes,” and the reinforcement sequence begins.
Next trial: Mum holds the hat once again in a hovering position about 5 cm above Lilly’s head. As soon as Lilly’s hands are placed on each side of the hat, Mum releases it, and the hat drops onto Lilly’s head. “Yes,” Mum says, and the reinforcement delivery chain begins again. In the following trial, Mum presents the hat at eye level. Lilly grasps the sides, Mum lets go, but Lilly lets it fall onto the table. Without saying anything, Mum re-presents the hat, returning to the previous, easier criterion, “yes,” reinforcement delivery chain commences. Then Mum attempts to advance again, presenting the hat at eye level: Lilly extends her arms forward, holding the sides of the hat, draws them back, and places the hat on her head.
Mum continues in this way—sometimes increasing the requirement, sometimes returning to a previously met criterion—systematically shaping the chain of picking up the hat and putting it on, without physical guidance. Within 10 minutes, the hat is presented on the table in various positions; each time, Lilly picks it up and places it on her head. For the first time, Lilly is learning the functional use of an object through her own initiated movement, adjusting her actions based on feedback. On the final trial of the day, Lilly places the hat on her head, but it slips to the floor. She bends down, picks it up, puts it back on her head, and pushes it in so it stays on. Mum unintentionally forgets to say “yes,” astonished by the new response. Lilly adjusts the visor to the front and looks at Mum expectantly, almost as if to ask, “Is this better?” Mum responds, “Yes, yes, yes, yeees!”
A tug of war
Over the summer, Mum and Lilly teach each other many skills, using the same shaping logic as in the hat example. Mum learns to break movements into components. For example, teaching Lilly to close a marker requires first picking it up with the tip up, then turning it horizontally, holding it steady, picking up the lid, and finally pushing the marker into the lid. What looked like a single action is, in fact, a chain of smaller connected movements. Mum thought this would be too difficult, so she decided to teach Lilly to open the marker instead, inductively figuring out the shaping steps. First, Mum presented the marker lid-up, holding most of the marker’s body in a closed fist to induce picking it up from the top. As soon as Lilly closed her fingers around the lid, Mum let go. This continued until Lilly began to apply a bit of force when picking up the marker from the lid (Mum simply held on a little tighter, as if they were tugging). “Yes.” Then Mum loosened the lid so that when Lilly pulled, it came off easily. Later, she closed it fully so Lilly had to pull harder to get the lid off. Finally, Mum covered the lid with her fist, evoking the grasp from the bottom of the marker with the opposite hand; when Mum let go, Lilly used the other hand to remove the lid. These were beautiful antecedent presentations.
Small steps forward, and sometimes back
Using the same shaping strategy, Lilly went on to learn many other functional object actions: opening an umbrella, putting on her wellies, stirring semi-liquid meals, and pouring water without spilling. None of these involved physical prompting—just careful antecedent arrangements to minimise errors, precise timing, a consistent marker (“yes”), a consistent reinforcement delivery sequence, and a clear end-of-consumption signal.
In most cases, Mum used what we could term backward shaping: arranging the starting point so that the distance between first contact with the object and the final criterion was minimal, as in the hat and marker examples. Sometimes, what appeared to be a single movement was broken into several smaller actions, with each shaped separately before being put all together.
When prompting falls short
Why shaping instead of graduated guidance or model prompts? Lilly had a long history of physical prompting and had never developed an imitation repertoire. For a modelled action to establish stimulus control over a movement, the learner must already be able to copy; without that repertoire, the model cannot function as a prompt. Physical prompting from start to finish had produced some topographies—like pulling down trousers to use the toilet or fastening a seatbelt—but in each case, the response began with the adult’s movement, not Lilly’s. Shaping ensured that each trial started with Lilly’s own initiated action, allowing feedback (or its temporary absence) to guide the refinement of her movements.
It’s about the process, not the response count
What mattered, in Lilly’s case, was not just how many responses could be “ticked off” on an item list, and presented on a cumulative graph, but the learning process that produced them. The most meaningful change is in how active her responding has become, the extent to which feedback, or its absence, selects changes in her movements. Direct prompting, especially physical, bypasses that process. The learner may eventually perform the response independently, but without learning how to learn it. Shaping builds the skill of adjusting one’s actions contingent on differential feedback, a repertoire that extends far beyond the acquisition of any single topography.
Shaping need not involve learning through trial and error—though that is a valuable skill in itself—but something more foundational: changing one’s response in contact with feedback as a skill in its own right. Effective shaping (and Kiki Yablon would say “modern shaping”) minimises errors by choosing the starting point precisely, shifting criteria in small, steady steps, and adapting moment to moment to the learner’s behaviour.
Breaking down imitation
Once about twenty actions were established this way, it was time to introduce imitation, where matching the model is the maintaining consequence. Because Lilly already had a repertoire of functional object actions initiated and refined through shaping, the first imitation skill was “Pick up what I pick up.” Three rubber coloured circles (about 10 cm in diameter) were placed in front of Mum and three in front of Lilly in mirror positions. To keep with the no-direct-prompting principle, Mum first shaped “place your hand in the circle that mirrors mine.” When Mum placed her hand on the middle circle, Lilly did the same; when Mum touched the right circle, Lilly touched her left.
Once Lilly could do this reliably, objects were placed in each circle, and the mirror-touching sequence was repeated, but now the item touched was picked up and placed in a box. Lilly did the same. Next, the mirror positions were removed so that the controlling dimension was object identity, not position. This established a conditional discrimination: Lilly responded differentially to which object Mum picked up.
Mum then replaced the neutral objects with those whose functional actions had already been shaped. The sequence became: watch and wait → “yes” → end of model action → go → “yes.” Once Lilly could select and use the matching object, Mum tested what might be called “true” imitation: reproducing both familiar and unfamiliar actions with both familiar and unfamiliar objects.
The question now was whether her shaped movement repertoire would allow novel combinations to emerge when modelled. Could Lilly observe a new action with a familiar object and reproduce the modelled behaviour instead of the previously shaped one? How about novel objects and familiar actions? And finally, unfamiliar objects and novel actions? She could.
Education is what survives when what has been learned has been forgotten
Skinner once wrote, “Education is what survives when what has been learned has been forgotten” (1964, New Scientist). In that article, he pointed out that schooling is often concerned with the acquisition of facts—verbal responses to specific verbal antecedents—with little attention to the skills required to acquire those facts, which are typically treated as by-products. Those facts, those fixed intraverbal relations, are subject to degradation over time if they are not consistently maintained—how many of us remember chemical formulas or historical dates? Yet we could learn them again if adequately motivated. What survives is the complex repertoire that allowed us to acquire those responses in the first place: the ability to learn novel responses. I believe that is what Skinner meant in his famous quote on education.
In ABA curriculum design, we often focus on establishing specific topographies within a particular response class. Programme sheets detail the prompts and the fading steps, but we devote increasingly little attention to the learning process itself: how the responses are acquired. Too often, we simply prompt and fade the prompt, bypassing the learner’s opportunity to contact the environmental contingencies that would otherwise shape the response. I interpret this to be the challenge that Skinner set us in 1964, and it is one we still have to fulfil: how do we conceptually define, and then establish, how to learn? What matters is not the topography of a single act, but how that act emerges. Shaping is hard. Not on the learner, but on the teacher. Shaping teaches the teacher how to teach: how to engineer the environment so that the response is initiated by the learner, and to be an active observer and listener, making the response more probable without giving the learner the response. When done well, shaping facilitates the emergence of a learner who is no longer waiting to be moved, but is constantly moving. And movement is behaviour.
.